{"id":"arxiv/2611.04448","url":"https://arxiv.org/abs/2611.04448","tags":["math.OC"],"score":0,"title":"Relative-Error Composite Federated Optimization with One Server-Side Proximal Step per Round","domain":"math.OC","source":"arxiv","authors":["Sirong Dai","Jianting Pan","Ming Yan"],"reviews":0,"subject":"math.OC","summary":"We consider deterministic composite federated optimization, in which distributed smooth client objectives are combined with a shared, possibly nonsmooth, regularizer. We propose Relative-Error Composite FedDyn (REC-FedDyn), an inexact method that supports partial participation, with one server-side proximal evaluation per communication round. REC-FedDyn employs a directly testable relative-error criterion that allows each client to independently control the accuracy of its local subproblem, without a server-side acceptance test. We establish almost-sure iterate convergence for strongly…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-11-03","reactions":{}}